Bad Quality in a Greek LPC Diphone voice

Hello everyone,

I have created a Greek LPC diphone voice but the prosody is really bad. The
phoneme junctures are ok, which means that the labeling and the segmentation is
ok. So i think that the pitchmark extraction is problematic. I have uploaded
some samples of the voice in the following link in order to check them out if
you have time and verify my assumption.

http://rapidshare.com/files/153562517/Eval.rar.html

Thank you very much in advance,

Chris Boubousis

===========================================
Boubousis Chris
Undergraduate Student

Speech Group
Telecom Laboratory
Department of Electronics & Computer Engineering
Technical University of Crete
Chania 73100, Greece

Re: Bad Quality in a Greek LPC Diphone voice

В Пнд, 13/10/2008 в 15:00 +0300, cmpoumpousis <at> telecom.tuc.gr пишет:
> Hello everyone,
> 
> I have created a Greek LPC diphone voice but the prosody is really bad. The
> phoneme junctures are ok, which means that the labeling and the segmentation is
> ok. So i think that the pitchmark extraction is problematic. I have uploaded
> some samples of the voice in the following link in order to check them out if
> you have time and verify my assumption.
> 
> http://rapidshare.com/files/153562517/Eval.rar.html
> 
> Thank you very much in advance,

Well, f0 is certainly wrong. Please provide a sample wav file from your
database, a corresponding pitchark file and also please provide an
utterance which you can get with 

(utt.save (SayText "some text"))

Building automatic LTS rule. Any guide for unicode?

Re: Building automatic LTS rule. Any guide for unicode?

Firoj Alam wrote:
> I am trying to build automatic LTS rule for "Bangla TTS". I have lexicon data in unicode format. The
automatic LTS developing tool which is available in festival does not work for unicode. But general
lts.ruleset can handle unicode by reading three byte as a [one unicode] character. Is there any way to
make/modify the system [lts_build], which can handle the unicode.
> 

Hi Firoj,

Yes it does work with unicode but you need to be a little careful at the 
first state.  However instead of having your initial lexicon as

( "LLLLLL" nil (p p p p p p p))

If you have

( ("L" "L" "L" "L" "L") nil (p p p p p p p)

Where "LLLLL" is a unicode word and "L" "L" "L" ... are the unicode 
split letters.

In the latest version of the LTS scripts in festvox/src/lts, the
setup option can take a third argument asis or utf8 which should make it 
do the right thing with utf8

$FESTVOXDIR/src/lts/build_lts setup yourlex.scm utf8

This has worked for lots of languages including Hindi/Devanagari.

Alan

> Regards
> Firoj
> 
> 
> Send instant messages to your online friends http://uk.messenger.yahoo.com 

Re: Building automatic LTS rule. Any guide for unicode?